Turbine blade defect detection method based on improved YOLOv8s

The performance and integrity of Aero-engine turbine blades are crucial for the normal operation of the engines. This study presents a real-time defect identification system for Aero-engine turbine blades utilizing an enhanced YOLOv8s architecture. Challenges like human dependency, hidden defect detection, and lack of real-time monitoring are addressed. The HDDSSPPF module substitutes the conventional Spatial Pyramid Pooling component to capture extended receptive field coverage. By implementing sequential dilated convolutions with differential expansion ratios, this architecture incorporates comprehensive contextual features and improves object boundary delineation accuracy. This structural enhancement significantly boosts the framework's capacity for holistic feature extraction compared to the standard SPPF configuration. Subsequently, the reparametrized ghost (RepGhost) bottleneck structure is integrated into the C2f module. Moreover, the bidirectional feature pyramid Network (BiFPN) replaces the Concat to enrich feature integration. To optimize training efficacy on complex detection cases, the MPDIoU metric (Minimum Point Distance Intersection over Union) was implemented as the objective function, specifically designed to strengthen feature representation for problematic instances. Experimental research was conducted on typical defects using a self-developed spacecraft blade dataset. The findings show that, in comparison to the original YOLOv8s, precision improves by 2.7% from 92.4%, and mAP (0.5) increases by 3.82–98.4%. This suggests that the proposed model enhances real-time detection performance for spacecraft blade surface defects.